Reckoning and Judgment

Brian Cantwell Smith introduced the reckoning–judgment distinction in The Promise of Artificial Intelligence (2019) as the organizing framework for evaluating what the deep-learning revolution had and had not accomplished. Reckoning is the manipulation of representations: symbols, vectors, tokens, pixels transformed according to rules or learned patterns into other representations. The transformation can be staggeringly sophisticated—it can defeat world champions at Go, fold proteins, draft legal briefs—but it remains, at its root, a transformation of representations. Judgment is something different in kind: the deliberate, committed, accountable disposition of a system that takes responsibility for getting the world right, not right relative to a training distribution but right relative to how things actually are. The two differ not in degree of sophistication but in the kind of relationship they involve to the world the representations are about. Smith drew on the older notion of practical wisdom—the seasoned discernment that knows what a situation calls for when no rule fully specifies it—and argued that no increase